Torque control of electric motor via current gives the advantage of simplicity and fast response over the complicated torque control of an internal combustion engine which may depend on several parameters ranging from fuel valve angle to gas pedal position and several delay factors. Although traction control system (TCS) for in-wheel-motor (IWM) configuration electric vehicles (EV) has advantages, the performance of the control system, as in most traction control cases, still depends on (1) accurate estimation of road-tire friction characteristics and (2) measurement of slip ratio requiring expensive sensors for obtaining wheel and chassis velocity. The main contribution of this work is design and integration of an acoustic road-type estimation system (ARTE), which significantly increases the robustness and reduces the cost of TCS in IWM configuration EVs. Unlike complicated and expensive sensor units, the system uses a simple data collection set-up including a low-cost cardioid microphone directed to vicinity of road-tire interface. The acoustic data is then reduced to features such as linear predictive, cepstrum and power spectrum coefficients. For robust estimation, only some of these coefficients are selected based on minimum intra-class variance and maximum inter-class distance criteria to train an artificial neural network (ANN) for classification. The road types can be grouped into: Asphalt, gravel, stone and snow with a correct classification rate of 91% for the test data. The predicted road-type is used to select the correct friction characteristic curve (mu-lambda) which helps calculating the appropriate torque command for the particular road-tire condition. The system has been evaluated in extensive simulations and the results show that extreme torque values are supressed stabilising the vehicle for several driving scenarios in a more energy-efficient and robust manner compared to previous systems.